skibidi sandbox 01 

Thanks! I might be able to do other basic shapes like spheres (e.g. dippin' dots). In the past I used rasterization for spheres but I think now I can ray-intersect some perfectly smooth spheres too. The emitter shape could also be changed to a star shape. But photorealistic ice cream could be a bit tough to achieve. Might need real-time meshing.

Yeah, I will try to get that added soon. Want to do some code reorganizing and cleanup this week.

@@GrantKot Awesome, looking forward to that! I had a real blast making dams and enclosed containers in the city demo. It goes a bit wild when particles have nowhere to go, but when poking a hole the pressure works pretty much as expected :)

The collision handling shown here is quite low cost, barely affecting the fps. Unfortunately it's not continuous collision detection so if particles move too fast they can "tunnel". When the blocks are joined just by edge, on the small toilets, the edge area is actually very thin. I have a few ways that I will be testing out to address it. I have the particles grouped in tiles, so for particle tiles that are next to difficult scenarios, I should be able to enable more advanced collision handling.

I'll add this soon, with prototyping it's too easy to sometimes just hard code in the domain size. But I'm going to be adding some UI now and better scene management.

Is it possible with floating point and multithreading? Because depending on the order even simple additions can return slightly different results? I haven't had much experience with netcode yet.

@@GrantKot If it's CPU simulated then the floating point part should be fine I think, unless you use fast math opt. I have no idea how your sim works, but regarding multi threading I guess you could either try noita style checkerboard updates if it's single-buffered, or use some kind of double buffering?

Ok, you know, maybe I'll just try it out and improve it over time :) There's also Parsec/Steam Remote Play style P2P streaming multiplayer. I was even thinking of running a larger cloud server for a large number of players and larger world.

The sim can be made deterministic regardless of threading; splitting the workload between threads need not change the semantics of the math as long as the order of dependent operations is preserved. But 'real' multiplayer support is always about latency reduction (latency from player input to simulation response), which means eschewing the authoritative server model - unless you are ok with interactions that have higher latency for players more distant from the server (which might be fine in this case; the force-grab tool you show here is probably fine with a lot of latency since there's inertia in the sim anyway)

I think I'm still a really big fan of the cloud streaming multiplayer model. With the Parsec tests, latency seemed ok. Because of the system requirements, for determinism they would have to run at the settings of the slowest computer in the group. One thing I want is instead of split screen, send separate video streams to the other players for their own viewports, render & encode at the bitrate that works best per connection (so slowest connection doesn't drag everything down). Once I get things better looking I want to get in touch with the Nvidia Geforce Now or Xcloud peeps to see if they can do something like that. I think these cloud gaming would also be more popular if they supported easy multiplayer like that.

Thanks, this is on a laptop 13900HX and 4080. Simulation running on the CPU, and I plan on adding to the renderer to fully utilize the GPU.

You can use linear interpolation which is 4 in 2d and 8 in 3d. Linear is fine for liquids/gas. Viscoelastics need at least the quadratic, because the gradient for the linear kernel is discontinuous, and tracking the deformation gradient becomes difficult and it will lose shape more quickly/oscillate as particles cross cell boundaries. You can have different material groups use different kernels.

Yeah, might need to set up a little ray trace when advancing particles to the block grid.